Today OpenAI announced Apps in ChatGPT, AgentKit and ChatKit. I’m super excited to try them out. I will document my findings here.
OpenAI AgentKit
AgentKit is OpenAI’s suite for building, evaluating, and shipping AI agents. It bundles:
- Agent Builder: visual workflow + evals for fast iteration and versioning
- ChatKit: drop-in, customizable chat UI for your product
- Connectors: unified data/tool integrations and governance
Why it matters:
- Faster prototyping: wire tools and data, iterate with trace/eval built-ins
- Production-ready UI: focus on your logic, not chat plumbing
- Works hosted or self-hosted: start simple, migrate when needed
OpenAI ChatKit
ChatGPT-like UI you can embed to build agent experiences quickly. It handles streaming, threads, and tool results so you can focus on agent logic instead of chat plumbing. You can point it at an OpenAI-hosted agent built in Agent Builder or at your own backend. Today there’s a Python reference backend for self-hosting; other runtimes should be doable — I might build one in Java.
Resources
- chatkit-js - re-usable chat UI
- ChatKit Starter Template - a NextJS app using chatkit-js
Agent Builder
Visual canvas to prototype, version, and evaluate agents. Supports preview runs, datasets and graders, and can deploy agents that ChatKit can talk to.